0

The link here describes a method for image classification using affinity propagation. I'm confused as to how they got the feature vectors, i.e, the data structure of the images, e.g, arrays?

Additionally, how would I accomplish this given that I can't use Places365 as it's custom data (audio spectrograms)?

Finally, how would I plot the images as they've done in the diagram?

user2793618
  • 305
  • 4
  • 10

1 Answers1

1

The images are passed through a neural network. The activations of neural network layer for an image is the feature vector. See https://keras.io/applications/ for examples.

Spectrograms can be treated like images.

Sometimes even when domain is very different, the neural network features can extract useful information that can help you with clustering/classification tasks.

dgumo
  • 1,838
  • 1
  • 14
  • 18
  • How did they plot the images, and what neural network should I choose? I'm really new to unsupervised learning. – user2793618 May 29 '19 at 18:31
  • 1
    It is a scatterplot with images as annotations. https://stackoverflow.com/questions/48180327/matplotlib-3d-scatter-plot-with-images-as-annotations – dgumo May 29 '19 at 18:43
  • 1
    You can use any well-known neural network. VGG16 or Inception trained on Imagenet may work for your use case – dgumo May 29 '19 at 18:44
  • is there a reason why they used a 365 dimensional vector? – user2793618 May 29 '19 at 20:26
  • 1
    Because they are using last layer of the network trained for Places 365 dataset. Last layer has 365 outputs leading to a 365-d vector. – dgumo May 29 '19 at 22:19
  • Ah ok. That's really self explanatory. So then, just to clarify, the feature vector is done through using a neural net to predict, i.e, vgg16's predict function? – user2793618 May 29 '19 at 22:26
  • Is there a way for me to get the corresponding image to the corresponding point or cluster for image annotations? So far I've only found that they've just used a labeled image as a representation for the point, not the image that belongs to the point. – user2793618 May 30 '19 at 20:57